Advanced crossplot analysis through physics-informed velocity prediction: a machine learning solution for reservoir characterization in the Guantao Formation of the Bohai Bay Basin, China

Abstract

Accurate prediction of shear wave velocity (Vs) is crucial for seismic interpretation, reservoir characterization, and geomechanical analysis in hydrocarbon exploration. This study introduces a physics-informed neural network (PINN) that integrates Kuster–Toksöz (KT) effective medium theory with data-driven learning to enhance shear wave velocity prediction and derive formation elastic parameters. The KT model provides the governing physical equation within the PINN framework, ensuring predictions adhere to established rock physics principles while leveraging neural network pattern recognition capabilities. The KT-PINN was validated using measured shear wave velocity from multiple formations in the basin, demonstrating superior performance (i.e., R 2 = 0.96) compared to the rock physics model (R 2 = 0.81) and standard neural networks (R 2 = 0.91). The results show that the KT-PINN model maintained physical consistency, enabling a reliable crossplot for hydrocarbon detection, with clear discrimination between fluid types in terms of their VpVs ratios (1.82-1.97 for oil versus 1.97-2.2 for water), and lambda-mu-rho analysis ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>λ</mml:mi> </mml:math> <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>ρ</mml:mi> </mml:math> - <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>μ</mml:mi> </mml:math> <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>ρ</mml:mi> </mml:math> ) (12-18 for oil and 18-21 GPa×g/cm³ for water-saturated rocks), and Poisson's ratio domains ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>ν</mml:mi> </mml:math> = 0.25-0.35 for oil-saturated and <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>ν</mml:mi> </mml:math> = 0.35-0.45 for water). In addition, physics-informed 3D seismic inversion further identified premium reservoir intervals with porosity &gt; 20% and oil saturation &gt; 60%, concentrated in channel sandstones around the CB323 and CB327 wells. These findings demonstrate the scalability of well-log predictions to field-scale reservoir models, establishing KT-PINN as a robust combination of rock physics and machine learning for hydrocarbon detection in geologically complex settings.